Predictive Index® Technical Overview

Todd C. Harris, Ph.D.

Director of Research: PI Worldwide

Visiting Professor: Department of Psychology
University of Massachusetts at Amherst

Allison J. Tracy, Ph.D.

Methodologist and Senior Research Scientist:

Wellesley Centers for Women
Wellesley College

Gwen G. Fisher, Ph.D.

Assistant Research Scientist:

Institute for Social Research
University of Michigan

16 Laurel Avenue, Wellesley Hills

Massachusetts 02481-7532
Telephone: +1 781 235 8872
Toll-Free in USA: 800 832 8884
Fax: +1 781 235 0959
www.PIworldwide.com

© PI Worldwide. All rights reserved.

Introduction

If you have ever heard of the phrase “the people make the place”, then you’ll
understand why the use of personality assessments in business and industry
continues to grow rapidly. Since approximately the late 1980’s, the academic
study of personality and the application of personality theory toward the solution
of key organizational challenges has undergone a marked renaissance. Interest
in personality has also expanded past traditional domains such as personnel
selection and hiring (Roberts & Hogan, 2001) to touch upon diverse areas such
as the influence of personality on team performance, leadership, organizational
culture and climate, entrepreneurship and innovation, stress and well-being, work
motivation, job satisfaction, and a host of others. Personality assessments tap
into each employee’s unique “behavioral DNA,” yielding key insights into people’s
individual drives, temperaments and motivations (e.g. why I might enjoy
analyzing financial statements for hours on end, while you’d prefer to be out of
the office developing personal relationships with customers.) Hundreds of
empirical research studies, conducted in a wide variety of settings, have
conclusively demonstrated the quantitative connection between personality and
job performance (Barrick & Mount, 1991; Barrick, Mount & Judge, 2001). Quite
simply, scores on well-developed personality measures are stable across time
and situations, and are useful predictors of behavior and job performance.

Personality Defined

Personality has been defined in many different ways by a variety of theorists.

Harvard’s Gordon W. Allport, in his seminal textbook on the field (Allport, 1937),
defined personality thusly:

“Personality is the dynamic organization within the individual of those

psychophysical systems that determine his unique adjustment to his
environment.”

Another pioneer in the field of personality, Raymond B. Cattell of the University of

Illinois viewed personality as (Cattell, 1945):

“That which permits a prediction of what a person will do in a given

situation. The goal of psychological research is to establish laws about
what different people will do in all kinds of social and general
environmental situations.”

MacKinnon (1944) suggested that personality has two conceptually distinct

definitions. Used one way, personality refers to the distinctive and unique
impression that one makes on others. This perspective refers to personality from
the viewpoint of the observer, and is functionally equivalent to a person’s
reputation. Used in a second way, personality refers to the structures inside of a
person that are useful in explaining why a person creates a particular impression
on others.

®
Predictive Index Technical Overview rev. 02/2014 2
This is personality from the perspective of the actor, concerned with how a
person perceives him or herself, and is functionally equivalent to a person’s
identity. This “dual definition” of personality is also espoused by more
contemporary personality researchers and theorists (e.g., Hogan, 2004).

The Five-Factor Model (FFM) of personality normally attributed to Tupes and

Christal (1961) suggests that many existing personality constructs can be
organized in terms of five broad traits: (1) Extroversion, (2) Agreeableness, (3)
Conscientiousness, (4) Emotional Stability, and (5) Openness.

The Growth of Personality Assessment

What factors have led to the increased usage of personality assessments around
the world? First, the U.S. population is increasingly diverse, as are the
populations of many other countries. Personnel selection systems that rely
solely or primarily on measures of cognitive ability significantly adversely affect
most protected groups, especially African Americans, Native Americans and
Hispanics. White people are often hired at a disproportionately high rate when
typical cognitive ability tests are the primary selection and screening tools. These
adverse impacts created substantial pressure on companies to find equally valid
but less discriminatory selection techniques. Research on personality variables
indicates that they have much less, and often no, adverse impact on members of
protected groups (Hough, 1998), a tremendous advantage when dealing with
increasingly heterogeneous customer, supplier and employee bases.

Second, the research findings from the U.S. Army’s Selection and Classification
Project (known as Project A), a multi-million dollar, seven-year research effort
clearly indicated that the Army could improve it’s forecasts of overall job
performance via the addition of personality assessments to its battery of
cognitive tests.

Third, the appearance of the Five-Factor Model (FFM) in the early 1960’s spurred
a large amount of academic research in the area of personality, and provided a
reasonably comprehensive yet parsimonious taxonomy to organize the
measurement of personality. The amount of academic research on personality
continues to increase, as evidenced by the 40 different personality-based papers
presented at the 2010 Annual Society for Industrial and Organizational
Psychology Conference, and the appearance of personality-themed articles in
leading journals such as the Journal of Applied Psychology.

Fourth, influential meta-analytic reviews such as those by Barrick and Mount

(1991) and Tett, Jackson, and Rothstein (1991), in which the relationships
between a variety of measures of personality and a variety of job performance
metrics were examined provided further empirical evidence for the predictive
validity of personality measures.

®
Predictive Index Technical Overview rev. 02/2014 3
Fifth, “real-world” views held by non-psychologists continue to hold that the
qualities and characteristics within individuals have a crucial impact on their
performance at work. For example, an examination of most job postings
indicates that employers frequently seek personality-driven factors such as social
skills, initiative, creativity, flexibility, etc. at least as often, if not more so, than
specific technical skills, experience or intellectual abilities. Thus, there appears to
be a widely held view that personality really matters.

The last factor that may be driving the increased importance and usage of
personality assessments in business and industry is the nature of work and the
economy itself. The world of work has changed more rapidly in the past 15 years
than in the prior 100 combined. Eight fundamental forces are shaping the 21st
Century workplace:

1. Increased technology, especially information technology.

2. Increased diversity and globalization.
3. Increased prevalence of collaborative, team-based work structures.
4. Increased concern for person-organization “fit.”
5. The importance of work/non-work balance.
6. The frequency and pace of change.
7. An increased emphasis on learning.
8. New expectations and definitions of leadership.

These factors do not operate independently, but rather accentuate and multiply
one another. For example, leaders who operate in today’s global environment
have to possess fundamentally different skill sets than their predecessors of prior
generations did. As a group, these conditions make the explicit consideration of
applicant and employee personality factors by organizations more critical than
ever before.

The Predictive Index®

The Predictive Index® (PI®) is a theory-based, self-report measurement of

normal, adult, work-related personality, and has been developed and validated
exclusively for use within occupational and organizational populations. The PI
employs a free-choice (as opposed to forced-choice) response format, in which
individuals are presented with two lists of descriptive adjectives, both containing
86 items, and are asked to endorse those which they feel describe them (the
“SELF” domain), and then those which they feel coincide with how they feel
others expect them to behave (the “SELF-CONCEPT” domain). Summing across
these two domains yields a third implied domain (the “SYNTHESIS”), which can
be interpreted as reflecting an employee’s observable behavior in the workplace.
The assessment is un-timed, generally takes approximately five to ten minutes to
complete, and is available in paper-and-pencil, desktop and internet formats.
The PI® has been in wide-spread commercial use since 1955, with minor
revisions to the assessment occurring in 1958, 1963, 1988 and 1992. These
minor revisions were undertaken to both improve the PI’s psychometric
properties and to insure that each of the individual items on the assessment
conformed to appropriate and contemporary language norms.
®
Predictive Index Technical Overview rev. 02/2014 4
The PI® is currently used by over 7,800 organizations across a wide variety of
industries and company sizes, including 51 companies listed in the 2009 Fortune
500, and 82 companies listed in the 2009 Fortune Global 500. Organizations
that use the PI® are located in 143 different countries, with approximately 30% of
PI® utilization occurring outside of North America. In 2010, over one million
people around the world completed the Predictive Index® assessment. The PI® is
used for a variety of talent management purposes, such as personnel selection,
executive on-boarding, leadership development, succession planning,
performance coaching, team-building, and organizational culture change among
others.

PI® Construct Definitions

The PI® measures four primary and fundamental personality constructs:

1. DOMINANCE: The degree to which an individual seeks to control his or
her environment. Individuals who score high on this dimension are
independent, assertive and self-confident. Individuals who score low on
this dimension are agreeable, cooperative and accommodating.
2. EXTROVERSION: The degree to which an individual seeks social
interaction with other people. Individuals who score high on this
dimension are outgoing, persuasive and socially-poised. Individuals who
score low on this dimension are serious, introspective and task-oriented.
3. PATIENCE: The degree to which an individual seeks consistency and
stability in his or her environment. Individuals who score high on this
dimension are patient, consistent and deliberate. Individuals who score
low on this dimension are fast-paced, urgent and intense.
4. FORMALITY: The degree to which an individual seeks to conform to
formal rules and structure. Individuals who score high on this dimension
are organized, precise and self-disciplined. Individuals who score low on
this dimension are informal, casual and uninhibited.

The PI® also measures two “secondary” personality constructs, which are derived
from a combination of each of the four “primary personality” constructs described
previously:
1. DECISION-MAKING: Measures how an individual processes information
and makes decisions. Individuals who score high on this dimension are
objective, logical and are primarily influenced by facts and data.
Individuals who score low on this dimension are subjective, intuitive and
are primarily influenced by feelings and emotions.
2. RESPONSE-LEVEL: Measures an individual’s overall responsiveness to
the environment, which is reflected in his or her energy, activity level and
stamina. Individuals who score high on this dimension have an enhanced
capacity to sustain activity and tolerate stress over longer periods of time.
Individuals who score low on this dimension have less of this capacity.

®
Predictive Index Technical Overview rev. 02/2014 5
The scoring of the Predictive Index® checklist produces a behavioral pattern with
three elements, known as the Self, the Self-Concept and the Synthesis. The Self
measures a person’s natural, basic and enduring personality. The Self Concept
measures the ways in which a person is trying to modify his or her behavior to
satisfy perceived environmental demands. Lastly, the Synthesis, which is a
combination of the Self and Self-Concept, measures the ways in which a person
typically behaves in his or her current environment.

Users of the Predictive Index® should be aware of the following: (1) as

demonstrated by our nearly 500 criterion-related validity studies, dating back to
September of 1976, the PI’s results are job-related and significantly predict job
performance across a wide range of jobs and performance criteria; (2) PI’s job
analysis tool, known as the PRO, can be used to independently establish the
behavioral requirements and benchmarks for a job; and (3) at any time, as part of
the service contract, a client can request us to perform, free of charge, a local
criterion-related validity study that uses their own employees and performance
measures, quantifiably demonstrating the connection between the PI® and job
performance in their workplace.

Please note that all research studies on the PI® conducted by PI Worldwide are
designed and executed in a manner that is consistent with the U.S. Federal
Regulations of the Equal Employment Opportunity Commission’s (EEOC)
Uniform Guidelines for The Development and Use of Personnel Selection
Procedures (1978). These guidelines are designed to provide a framework for
determining the proper use of tests and other selection procedures, as well as
preventing discriminatory employment practices. Research on the PI® is also
conducted in accordance with the guidelines for test development established by
professional organizations such as the American Psychological Association
(APA), and the Society for Industrial and Organizational Psychology (SIOP). This
research indicates that: (a) traits measured by the Predictive Index® add
incremental validity to the prediction of job performance, and (b) use of the
Predictive Index® within a personnel selection system does not produce adverse
impact.

At present, PI Worldwide conducts and supports scientific research in the

following areas:

The Reliability of PI®

Reliability refers to the consistency or stability of a measure (Nunnally &

Bernstein, 1994). If the concept being measured is assumed to be consistent,
such as a personality trait, then the measure should yield similar results if the
same person responds to it a number of times. If the concept being measured is
assumed to be inconsistent, such as mood, then the measure should yield
dissimilar results if the same person responds to it a number of times.
One way to estimate reliability is by computing the measure’s “test-retest
reliability”. Test-retest reliability is perhaps the easiest assessment of a
measuring device’s reliability to conceptualize and understand. Using the same
group of people, a construct is measured at two separate points in time and then
®
Predictive Index Technical Overview rev. 02/2014 6
the two sets of scores are compared. This technique yields a correlation often
known as the coefficient of stability, because it reflects the stability of test scores
over time. If the measure under study is reliable, people will have scores that are
similar across trials. Note that the shorter the time interval between
administrations of the test (e.g. two weeks versus three months), the higher will
be the test-retest coefficient.

The test-retest reliability of the PI® was first examined in a 1983 study by Perry &
Lavori, who reported the following coefficient of stability values:

Table 1: PI® Test-retest Reliability Results from Perry-Lavori

Sample Size N = 85 N = 87 N = 86
Time Interval Three Two Years Four Years
Months – – Four – Eight
Two Years Years Years
Self A (Dominance) .71 .53 .49
Self B (Extroversion) .67 .57 .56
Self C (Patience) .54 .58 .52
Self D (Formality) .70 .61 .56
Self-Concept A (Dominance) .56 .45 .47
Self-Concept B (Extroversion) .48 .47 .50
Self-Concept C (Patience) .60 .51 .44
Self-Concept D (Formality) .66 .46 .47
Synthesis A (Dominance) .67 .55 .52
Synthesis B (Extroversion) .60 .55 .56
Synthesis C (Patience) .61 .60 .49
Synthesis D (Formality) .69 .59 .54
Synthesis M (Response Level) .60 .57 .51

A 1999 study by Everton reported the following coefficient of stability values:

Table 2: PI® Test-retest Reliability Results from Everton

®
Predictive Index Technical Overview rev. 02/2014 7
 Sample Size N = 77 N = 58
Time Interval Two Weeks Six Months
Self A (Dominance) .80 .75
Self B (Extroversion) .71 .80
Self C (Patience) .76 .71
Self D (Formality) .80 .57
Self-Concept A (Dominance) .68 .65
Self-Concept B (Extroversion) .61 .73
Self-Concept C (Patience) .71 .48
Self-Concept D (Formality) .56 .57
Synthesis A (Dominance) .84 .76
Synthesis B (Extroversion) .69 .82
Synthesis C (Patience) .83 .62
Synthesis D (Formality) .74 .60
Synthesis E (Decision-Making) .71 .58
Synthesis M (Response-Level) .81 .71

2009 and 2011 studies by Harris, utilizing samples of 61 and 44 undergraduate

students respectively, reported the following coefficient of stability values over a
two-week interval:

Table 3: PI® Test-retest Reliability Results from Harris

PI® Factor Stability Coefficient: Stability Coefficient:

2009 Study 2011 Study
Self A (Dominance) .84 .79
Self B (Extroversion) .77 .81
Self C (Patience) .75 .70
Self D (Formality) .83 .87
Self-Concept A (Dominance) .72 .70
Self-Concept B (Extroversion) .69 .63
Self-Concept C (Patience) .65 .73
Self-Concept D (Formality) .77 .74
Synthesis A (Dominance) .81 .78
Synthesis B (Extroversion) .82 .84
Synthesis C (Patience) .78 .78
Synthesis D (Formality) .75 .68
Synthesis E (Decision-Making) .78 .80
Synthesis M (Response-Level) .70 .73

Taken as a whole, these results indicate that the PI® demonstrates acceptable
levels of test-retest reliability.

A second way to estimate reliability is by computing the measure’s “internal

consistency reliability.” This is accomplished by determining whether the
individual items on the assessment intended to measure the same construct
(such as Dominance) are mathematically related. Internal consistency methods
®
Predictive Index Technical Overview rev. 02/2014 8
estimate the reliability of a test based solely on the number of items within the
test and the average inter-correlation among those items. The internal
consistency reliability of the PI® has been examined in three different studies.
The average internal consistency reliability of PI® Factors across these studies is
.85, with a range of .82 to .87. Although estimates vary, the lower boundary for
the acceptability of internal consistency reliability is often taken as .70. There
may also be an upper boundary of acceptability as well, perhaps .90 or above, as
that may signal measurement redundancy across some of the items.

Note that the last exploration of the internal consistency of PI® constructs was
undertaken in 2009, using a randomly selected sample of 3,991 applicants to and
employees from a variety of US businesses in a variety of industries: Information
Technology, Financial Institutions, Surface Transportation, Air Transportation,
Manufacturers, Wholesalers, Retail, Hospitality, and Restaurants. The sample
contained individuals from all levels of their respective organizations, and
included a variety of jobs, such as warehouse workers, tellers, accountants,
pilots, inside and outside sales, and vice presidents.

The following alpha coefficients (Cronbach, 1951) were obtained in this study:
Self Concept (Factor A=0.78, Factor B=0.87, Factor C=0.77 & Factor D=0.84)
and Self domains (Factor A=0.80, Factor B=0.88, Factor C=0.82 & Factor
D=0.86). Note that in 1996 study based on data from a sample of 1,411
individuals, internal consistency reliability estimates for Self Factors ranged from
0.79 to 0.87 (Wolman & Pratt, 1996).

PI® Measurement and Scaling Structure

The PI® assumes that a single personality dimension is being tapped by each PI ®
factor scale. For example, Factor A is expected to assess the broad personality
trait of dominance. Nuances of this trait are captured in the individual adjectives
and summed to broadly define dominance. This assumption of unidimensionality
within each factor is examined empirically in this evaluation.

According to classical test theory, the variability in an item set can be separated
into shared, or “true score” variance and unique, or “measurement error”
variance (see Allen & Yen, 2002). In other words, differences between
individuals’ responses to an adjective can be attributed to actual differences in
the personality dimension it is designed to measure (“true score”) and to
unrelated influences (“measurement error”). For instance, the word "popular" can
evoke images of specific attributes of an individual's past schoolmates in addition
to more abstract notions of popularity associated with extroversion. The degree
to which an item is influenced by factors other than that intended is referred to as
measurement error. All psychological assessment instruments contain some
proportion of measurement error.

In this evaluation, individual items are evaluated in an effort to discover the

extent to which they contribute true score variance to the factor score estimates.

®
Predictive Index Technical Overview rev. 02/2014 9
Note that the adjectives reflecting each PI® factor are words of degree - not true
opposites. For example, a peaceful person can be somewhat belligerent. Item
Response Theory (IRT) scaling and other measurement models assume that
items are ordered according to their ability to detect differences in degree at
different points along the underlying personality continuum. In other words, one
item may be endorsed at very low levels of a personality dimension and, as the
degree of the dimension increases, another item may also be endorsed.

Because psychological dimensions cannot be directly observed but must be

inferred from measurement characteristics of the items in the assessment
instrument, changes in the measurement model (item changes, shifts in the
proportion of measurement error, etc.) imply changes in the interpretation of the
factor. Meaningful comparison of factors requires the assumption that factor
scores have the same meaning in both the Self Concept and Self domains. In
other words, scaling must be identical across domains if one is to say, for
instance, that an individual perceives himself to be less formal than he feels
others expect him to be. Since factor scores are composites of item sets, this
assumption implies that each item reflects a personality dimension in the same
way in the two domains and has a similar proportion of measurement error.

The measurement properties of the items underlying the construction of factor

scores allow us to explicitly test the assumption of equivalence of scale across
domains.

A key feature of the PI® is that factor scores can be used to construct an
individual’s personality profile, or factor pattern. Because the relative intensity
levels of the four personality dimensions measured by Factors A through D are
essential to the interpretation of personality profiles, there is an implied
assumption of comparability of the metric of factor scores. In other words,
personality tendencies representing a similar degree of intensity should be
plotted at similar points on the PI® factor plotting template. Only by making this
assumption regarding the scaling of the factors can one say, for example, that an
individual has a higher tendency toward dominance than extroversion (High
A/Low B).

Principal component analysis, a special case of exploratory factor analysis, can

be used to evaluate the assumption of unidimensionality in the sets of items
comprising each PI® Factor. Utilizing the same sample of 3,991 applicants and
job incumbents referenced above, eigenvalue decomposition was used to create
a scree plot for each item set, which shed light on the optimal number of
components (factors) that can be extracted. Scree plots obtained from models for
each PI® factor showed evidence of unidimensionality. As can be seen in the
Appendix of this document, Figures 1 (Self-Concept) and 2 (Self), there is a
decidedly sharp drop-off after the first eigenvalue for each of the Factors A
through D, followed by a relatively flat slope.

This implies that the most parsimonious solution describing the item set is one
with a single underlying dimension. Additional dimensions lend only negligible

®
Predictive Index Technical Overview rev. 02/2014 10
explanatory value to the solution. This examination supports the assumption that
a single personality construct is described by adjectives in each item set.

Comparative Fit Index (CFI) values, which is a standardized scale assessing

overall model fit, for Self Factors A, B, C and D were CFI = .92, .96, .95 and .90
respectively. The Root Mean Square Error of Approximation (RMSEA), which
summarizes the degree of local fit, or the consistency of fit throughout all areas of
the model (Rigdon, 1996) = 0.06.

Individual item loadings in Factor A ranged from 0.49 to 0.76 in the Self-Concept
domain (mean=0.64) and from 0.52 to 0.78 in the Self domain (mean=0.64). In
Factor B, the loadings in the Self-Concept domain ranged from 0.51 to 0.75
(mean=0.64) and in the Self domain, loadings ranged from 0.56 to 0.74
(mean=0.67). Two items had loadings lower than 0.40: “Amiable” (Factor C, Self
Concept and Self domains) and “Passive” (Factor D, Self-Concept domain). The
remaining loadings in Factor C ranged from 0.60 to 0.78 in the Self-Concept
domain (mean=0.66) and from 0.57 to 0.72 in the Self domain (mean of 0.66).
Remaining loadings for Factor D ranged from 0.44 to 0.71 (mean=0.57) in the
Self-Concept domain and from 0.45 to 0.70 (mean=0.61) in the Self domain.

In 2003, an examination of the factor structure of the Swedish version of the PI ®

was undertaken by Dr. Lennart Sjoberg of the Stockholm School of Economics
(Sjoberg, 2003), starting with two factors and going up to ten. The results of this
analysis indicated only small drops in eigenvalue size after the fourth factor,
giving support to the notion of a maximum of four factors.

The Validity of the PI®

While reliability refers to the consistency of a measure, validity refers to the

accuracy of a measure. A measure is valid if it actually measures what it
purports to measure. The validity of a measure can be assessed in a number of
ways.

Construct validity is demonstrated when a measure is statistically compared with

another measure of similar and/or different concepts. To be successful, the
comparison measure must have been soundly constructed and be generally
accepted. Such research on the PI® has been conducted twice. Both of these
studies compared the PI® with Raymond Cattell’s 16PF (The 16PF is a well-
respected and well-researched personality assessment). A construct validity
study involves looking at patterns of correlations. Correlations are mathematical
measures that can identify the presence and strength of the relationship between
two variables.

A pattern should emerge that meets the following expectations: Factors that are
defined in a similar way by both the PI® and the 16PF should prove to be very
similar statistically (e.g., PI’s extroversion factor and the 16PF extroversion
®
Predictive Index Technical Overview rev. 02/2014 11
factor), and factors that are defined in a dissimilar way on both the PI ® and the
16PF should prove to be mathematically unrelated (e.g., PI’s extroversion factor
and the 16PF emotional stability factor).

In both of these studies, the PI® successfully demonstrated construct validity: the
relationships you would intuitively expect should be related were mathematically
related, and the relationships you would intuitively not expect should be related
were mathematically unrelated. For example, the correlation between PI’s
Dominance factor and the 16PF’s Independence factor was .47 (p<.01), between
PI’s Extroversion factor and 16PF’s Extroversion factor was .34 (p<.01) and
between PI’s Patience factor and 16PF’s Tension Factor was .35 (p<.01),
indicating that respondents who had relatively low Patience scores as measured
by the PI® tended to also be tense, impatient and driven as measured by the
16PF.

A 2010 pilot study (N = 186) comparing the relationship between the Predictive
Index® and analogous constructs measured by the NEO PI-R, discovered the
following correlations, all in the hypothesized direction and all significant at the
p<.01 level:

 PI® Factor A (DOMINANCE) and NEO PI-R “AGREEABLENESS”: r = -.61

 PI® Factor B (EXTROVERSION) and NEO PI-R “EXTROVERSION”: r = .63
 PI® Factor D (FORMALITY) and NEO PI-R “CONSCIENTIOUSNESS”: r = .61

Criterion-related validity is demonstrated when a measure is statistically

compared with behaviors it claims to predict. We say that the PI® is related to,
and can predict, behaviors in the workplace. Criterion-related validity studies
objectively show whether these relationships exist, and if so, they show the
nature of these relationships.

The PI® has been investigated in nearly 500 concurrent (in which data for the
predictor and criteria are collected at the same time) and predictive (in which
there is some time-lapse between when the data for the predictor and criteria are
collected) criterion-related validity studies since September of 1976, for a variety
of jobs, in a variety of industries, in a variety of countries and utilizing a wide
range of job performance metrics, such as tenure, turnover, sales, and customer
satisfaction. This body of evidence supports the fact that the PI® is indeed
consistently related to important workplace outcomes, with studies typically
yielding uncorrected correlations between PI® Factors and individual job
performance criteria in the .20 to .40 range. These correlations indicate that the
PI® can be an effective predictor of workplace performance.

Meta-Analytic Results for the Predictive Index®

Meta-analysis is a statistical procedure designed to combine the results of many

individual, independently conducted empirical studies into a single result or
®
Predictive Index Technical Overview rev. 02/2014 12
 outcome (Hunter & Schmidt, 1990; Raju et al., 1991). The logic underlying meta-
analysis is that researchers can arrive at a more accurate or “truer” conclusion
about a particular topic via combining or aggregating the findings of a wide cross-
section of many studies that address the topic, instead of relying on the results
obtained in a single study. In 2008, a PI® meta-analysis examining the most
recent 57 predictive and concurrent criterion-related validity studies was
conducted. These 57 studies were conducted over a time period ranging from
April, 2003 to February, 2008. Cumulative sample size was 5,765 (Mean N =
101.1; Median N = 74.0; Maximum N = 431; Minimum N = 32). The studies
comprising the meta-analysis were conducted on behalf of 51 different PI® clients
(including 21 Fortune 500 clients), drawn from 20 different industry classifications
(e.g. “financial services”) and 15 different occupational classifications (e.g.
“customer service”). 14 of the 57 studies were conducted with managerial-level
jobs, with the remainder focused on individual-contributor roles such as bank
tellers, customer service associates, truck drivers, machine operators, etc.

Meta-analytic results were as follows:

Table 4: P I®- Job Performance Correlations:

PI® Factor Mean Standard Deviation High Low
Self A (Dominance) .17 .09 .49 .02
Self B (Extroversion) .18 .10 .50 .00
Self C (Patience) .18 .10 .47 .00
Self D (Formality) .19 .10 .48 .03
Synthesis E (Decision-Making) .18 .10 .50 .00
Synthesis M (Response-Level) .18 .12 .51 .01

Table 5: PI® - Job Tenure Correlations:

PI® Factor Mean Standard Deviation High Low

Self A (Dominance) .13 .08 .35 .01
Self B (Extroversion) .15 .11 .47 .01
Self C (Patience) .14 .08 .41 .02
Self D (Formality) .14 .10 .39 .01
Synthesis E (Decision-Making) .11 .08 .33 .01
Synthesis M (Response-Level) .14 .09 .37 .02

A group of notes about the meta-analytic methodology employed in this meta-

analytic project warrants mention:

All of the above results were weighted by the respective sample sizes of
the original studies.
®
Predictive Index Technical Overview rev. 02/2014 13
 The correlations reported above have not been corrected for unreliability,
a fact that in all likelihood somewhat reduced the PI® – job performance
and PI® – tenure correlations reported here.
Both statistically significant (i.e. p<.05) and non-significant results were
included in the meta-analysis.
The correlations reported above did not differ significantly by either
industry or occupational classification. Additionally, the strength of the
relationships between the PI® and job performance and job tenure was the
same for managerial versus non-managerial jobs.
A wide variety of job performance criteria were included in the meta-
analysis, including objective production data (e.g. average units sold per
month), personnel data (e.g. number of unexcused absences over an
eight-month period) and judgmental data (e.g. supervisor ratings of work
quality). This may at least partially explain the variability of PI® – job
performance correlations reported in the individual criterion-related validity
studies.

The Impact of Demographic Variables on PI® Scores

We have performed research to determine whether the Predictive Index®

discriminates against protected classes. In a report written by Dr. Richard
Wolman of Harvard University, the PI® was analyzed to determine whether men
and women tended to score differently on the PI®, and whether African-
Americans, Hispanics, and Caucasians tended to score differently on the PI ®.
His analyses showed that neither gender nor race was significantly related to PI®
scores.

In a more recent study by Dr. Todd Harris, the PI® was analyzed to determine
whether the PI® produces adverse impact based on age. The study showed that
for all PI® factors, people over age 40 (the protected class) had PI® patterns that
were no different than people under age 40, confirming similar findings initially
obtained by Dr. Wendi Everton in 1998.

In a 2008 investigation of the impact of gender and race on PI® scores in a

sample of 347 employees in the banking industry, covering a variety of jobs (e.g.
teller, branch manager, loan officer, etc.), gender and race accounted for less
than 2% and 3% of the variability, on average, in PI® Factor scores. The results
of this body of research indicate that the PI® is age-, gender- and race-neutral,
and we believe that the inclusion of a well-validated personality assessment such
as the PI® in a company’s personnel selection system may lead to a more
demographically diverse workforce.

Note that there is no evidence to indicate that the inclusion of the PI® in a
company’s personnel selection system, either in a compensatory or “multiple-
hurdle” selection model, results in adverse impact against any protected class.
When examining adverse impact via the four-fifths rule, Adverse Impact (AI)
Ratios for large-scale studies of the PI® typically range from a low of .86 to a high
of 1.25. When examining the issue of adverse impact via the “two Standard
®
Predictive Index Technical Overview rev. 02/2014 14
Deviations” rule which we believe can be a more defensible and appropriate
approach in situations in which there are low sample sizes in either the applicant
or selected pool, we again see no evidence of the PI® producing adverse impact.
For example, in a recent multi-year study of bank tellers for a Fortune 1000
financial services company, a Z-score comparison of the selection ratios for
blacks and whites was Z = 1.51. In a second study of retail store managers, a Z-
score comparison of the selection ratios for males and females was 1.20. When
requested by the client, the PI Worldwide Research Department will partner with
PI® clients to examine adverse impact and other selection-related statistics,
incorporating both PI® and non-PI® predictors, and make any appropriate
recommendations. Custom research studies, including pass-rate and adverse
impact analyses, can also be conducted for PI® clients, designed in full
accordance with Uniform Guidelines and other professional standards.

Cross-Cultural Research on the PI®

The PI® has now been translated into 63 different languages, has been used
globally since 1958, and is seeing strong growth in Asia and India. We continue
to take steps to insure the PI’s validity in non-U.S. markets. For example, over
the past five years, we have conducted criterion-related validity studies involving
employees from China, Canada, India, Britain, Spain, Portugal, Germany,
Hungary, Russia, Australia and the Netherlands. The results of these
international job validity efforts demonstrate the same quantitative connections
between the PI® and job performance that our U.S.-based studies have shown.

Recent studies have also documented the PI’s internal consistency reliability,
freedom from bias, and construct validity in non-U.S. markets. For example, in
2000 and 2003 in studies of the Swedish version of the PI® checklist, Dr. Lennart
Sjoberg of the Stockholm School of Economics demonstrated that the PI ® had
acceptable reliability, validity and internal factor structure. The report “The
Construct Validity of the Predictive Index® in a South African Sample” (1996)
demonstrated PI’s construct validity as compared to the well-regarded 16PF
personality assessment.

The 2006 report entitled “The Impact of Race and Gender on PI® Scores in a
Large South African Sample” indicated that PI® scores were not significantly
impacted by race or gender in 1,620 employees from South Africa. Three
additional reports, published in 2007, demonstrated that the PI® is neither
impacted by age or nor gender in samples of employees drawn from the U.K.,
India or Romania.

A March 2011 study of 197 Predictive Index® profiles drawn from Poland (128
males; 61 females; 8 with sex data not available), covering a time period from
November, 2004 to February, 2011, and a range of occupational classifications,
found no statistically significant (p<.05) relationships between sex and any
personality construct measured by the Predictive Index®.

The Impact of Differing Modes of Administration on PI® Scores

®
Predictive Index Technical Overview rev. 02/2014 15
The results of our research indicate that an individual’s PI® scores are not
significantly impacted by mode of administration (i.e. paper-and-pencil vs.
desktop computer-administered vs. internet). For example, a 2007 project used
a stratified random sample of 15,000 PI’s selected from our client database and
covering a wide range of industries, companies, occupations and jobs. One third
of this sample had taken the PI® via paper-and-pencil (N=5,000), one third of the
sample had taken the PI® via an electronic kiosk/desktop mode (N=5,000) and
one third had taken the PI® via the internet (N=5,000). For each individual PI®
Factor, mode of administration accounted for much less than 1% of the variability
in individual scores on that Factor.

Additionally, the shapes of the distributions for all PI® Factors were consistent
across each of the modes of administration, being normally distributed (Skew
and Kurtosis values <1.0) with equivalent means and medians for all PI® Factors
at the Self, Self Concept and Synthesis levels of analysis. Additionally, the
results of factor-analytic work indicated that PI® factor structure was invariant
across these three modes of administration.

The Relationship between the PI® and General Cognitive Ability

When personality variables that are correlated with a job performance criterion
are added to a battery of other personnel selection instruments measuring
constructs such as general cognitive ability, psychomotor abilities, technical
skills, mechanical aptitude and vocational interests, predictive validity is typically
increased because personality variables are essentially uncorrelated with these
other potential predictors of job performance (Judge, Higgins, Thoresen &
Barrick, 1999). For example, in samples ranging in size from 7,188 to 8,547,
Hough, Kamp and Ashworth (1993) reported median correlations between
personality variables and measures of cognitively ability of .04. On a conceptual
level, personality and intelligence are typically viewed as separate constructs.
Intelligence has traditionally reflected the “can do” aspect of an individual; in
effect, the employee can do the job because he or she possesses an adequate
level of intelligence to do so. Personality has traditionally reflected the “will do”
aspect of the individual; in effect, the employee will do the job because he or she
has the motivation or temperament to do so. Personality and intelligence can
both be predictive of job performance, each in their own way. For example,
employees comparably more dependent on personality factors might achieve job
success via being dependable, conscientious, helpful, team-oriented and so on.
Employees comparably more dependent on intelligence might achieve the same
level of job success by being excellent problem-solvers, synthesizers of
information, and analyzers of data.

Our research has indicated that the PI® is unrelated to general cognitive ability.
In four recent studies in which this relationship was examined (aggregate
N=915), each of which utilized samples from different clients, industries and jobs,
the median absolute correlation between individual PI® Factors and a variety of
measures of general cognitive ability was .06 (Mean = .07; SD = 0.6).

®
Predictive Index Technical Overview rev. 02/2014 16
 Table 6: PI® – Cognitive Ability Correlations:

PI® Factor Study One Study Two Study Three Study Four Study Five
2002 2004 2005 2008 2010
N = 82 N = 172 N = 182 N = 178 N = 301
Self A (Dominance) .14 .19 .05 -.02 .09
Self B (Extroversion) -.19 .05 -.02 -.06 -.07
Self C (Patience) .05 -.18 .01 .09 .12
Self D (Formality) .06 -.17 -.03 .02 .15
Synthesis E .03 -.07 .04 .00 .02
(Decision-Making)
Synthesis M -.01 .08 .01 .04 .06
(Response-Level)

PI Worldwide Academic Collaborations

The Research Department at PI Worldwide welcomes the opportunity to

collaborate with academic researchers across disciplines. Here is a sampling of
some of the research projects that utilized the Predictive Index® in 2010:

Institution Department Research Topic

Indiana University Management The Role of Personality in
Volunteerism Choices and
®
Predictive Index Technical Overview rev. 02/2014 17
 Satisfaction
Eastern Kentucky University Psychology Administrators’ Personality
Traits and Nursing Home
Performance
Johnson and Wales University Hospitality The Relationship between
Personality Factors and
Student Attitudes
University of Sao Paulo (Brazil) Management Six Sigma and the Project
Portfolio Management
Process (PPMP): Predictors
of Success
Curtin University (Australia) Center for The Impact of Personality
Research on Aging on Staff Retention in the
Community Care
Environment
Capella University Education At a Distance: A
Comparative Study of
Distance Delivery
Modalities for Ph.D. Nursing
Students
University of Zaragoza (Spain) Psychology Role Evolution Among
Managers

®
Predictive Index Technical Overview rev. 02/2014 18
 References

Allen, M.J., & Yen, W. M. (2002). Introduction to Measurement Theory. Long

Grove, IL: Waveland Press.
Allport, G.W. (1937). Personality: A psychological interpretation. New York: Holt.

Barrick, M.R., & Mount, M.K. (1991). The Big Five personality dimensions in job
performance: A meta-analysis. Personnel Psychology, 44, 1-26.

Barrick, M.R., Mount, M.K., & Judge, T.A. (2001). Personality and performance
at the beginning of the new millennium: What do we know and where do we go
next? International Journal of Selection and Assessment, 9, 9-30.

Cattell, R.B. (1945). The principal trait clusters for describing personality.
Psychological Bulletin, 42(3), 129-161.

Cronbach, L.J. (1951). Coefficient alpha and the internal structure of tests.
Psychometrika, 16, 297 – 334.

Everton, W. (1999). A further investigation of the construct validity of the

Predictive Index®. (Validity Report No. 1999.11.01). Wellesley Hills, MA:
Praendex, Inc.

Hogan, R. (2004). Personality psychology for organizational researchers. In B.

Schneider & D.B. Smith (Eds.), Personality and organizations (pp. 1-23).
Mahwah, N.J.: LEA.

Hough, L.M., Kamp, J. & Ashworth, S.D. (1993). Development of Project A

temperament inventory: Assessment of background and life experiences (ABLE)
(Institute Report No. 259). Minneapolis, MN: Personnel Decisions Research
Institutes.

Hough, L.M., Ones, D.S., & Viswesvaran, C. (1998, April). Personality correlates
of managerial performance constructs. In R.C. Page (Chair), Personality
determinants of managerial potential performance, progression and ascendancy.
Symposium conducted at the 13th annual conference of the Society for Industrial
and Organizational Psychology, Dallas, TX.

Hunter, J.E., & Schmidt, F.L. (1990). Method of meta-analysis: Correcting error
and bias in research findings. Newbury Park, CA: Sage.

Judge, T.A., Higgins, C.A., Thoresen, C.J. & Barrick, M.R. (1999). The big five
personality traits, general mental ability, and career success across the life span.
Personnel Psychology, 52, 621-652.

MacKinnon, D.W. (1944). The structure of personality. In J. McV. Hunt (Ed.),

Personality and the behavior disorders (Vol. 1, pp. 4-43). New York: Ronald.

®
Predictive Index Technical Overview rev. 02/2014 19
Nunnally, J.C., & Bernstein, I.H. (1994). Psychometric theory (3rd ed.) New York:
McGraw-Hill.

Perry, J.C., & Lavori, P.W. (1983). The Predictive Index®: A report on reliability
and construct validity. (Validity Report No. 1983.09.01). Wellesley Hills, MA:
Praendex, Inc.

Raju, N.S., Burke, M.J., Normand, J. & Langlois, G.M. (1991). A new meta-
analytic approach. Journal of Applied Psychology, 76, 432-446.

Rigdon, E.E (1996). CFI versus RMSEA: a comparison of two fit indexes for
structural equation modeling. Structural Equation Modeling, 3, 369–379.

Roberts, B.W. & Hogan, R. (Eds.) (2001). Personality psychology and the
workplace. Washington, D.C. : American Psychological Association.

Sjoberg, L. (2003). Psychometric properties of the Swedish version of the

Predictive Index® Checklist. Stockholm, Sweden. Department of Economic
Psychology, Stockholm School of Economics.

Tett, R.P., Jackson, D.N., & Rothstein, M. (1991). Personality measures as

predictors of job performance: A meta-analytic review. Personnel Psychology,
44, 703-742.

Tupes, E.C., & Christal, R.E. (1961). Recurrent personality factors based on trait
ratings. Journal of Personality, 60, 225-251.

Wolman, R.N., & Pratt, J.H. (1996). A Normative Reliability Investigation of the
Predictive Index® Organization Survey Checklist. Wellesley Hills, MA: Praendex
Incorporated.

®
Predictive Index Technical Overview rev. 02/2014 20
 Appendix

7
6
5
Factor A
4 Factor B
3 Factor C
Factor D
2
1
0
1 2 3 4 5 6 7 8 9 10
Figure 1. Scree plots of the eigenvalues for the Self Concept domain (Form IV Evaluation Study,
subsample #1: n=3,991).

7
6
5
Factor A
4 Factor B
3 Factor C
Factor D
2
1
0
1 2 3 4 5 6 7 8 9 10

Figure 2. Scree plots of the eigenvalues for the Self domain (Form IV Evaluation Study,
subsample #1: n=3,991).

®
Predictive Index Technical Overview rev. 02/2014 21